[sejarzo]

Quote:

Originally Posted by HFat /img/forum/go_quote.gif This would mean that 0.5 V would be represented by 2^14 and 2^22 respectively, that 0.25V would be represented by 2^13 and 2^21 respectively and so on. So the "steps" would have exactly the same "size" until the 16 bits are exhausted, around the microvolt range. Therefore, the added "resolution" of 24 bit would be below the noise floor in most cases.

Not really, seems to me that you are equating the resolution of a 0.25V change quantified from 2^22 to 2^21 as being equivalent to it being quantified between 2^14 to 2^13, and that definitely isn't so. I might be misunderstanding you, or it just might be a semantic difference, though.

Punch it into a calculator and work it out for yourself. A 24 bit system does indeed provide much greater resolution of amplitude, and it really cannot be any other way.

As you note, the steps are indeed a constant factor of 2--they have to be, otherwise the system wouldn't be linear! But there are more bits available to resolve each 6 dB step in amplitude in the 24 bit system, so the added resolution is available throughout the entire dynamic range....not just at the lower end!

It is important to note that the Meridian paper does not specifically claim, as far as I could tell, that dithered quantization provides strictly infinite resolution--what I saw was a statement that it provides "effectively infinite resolution".

Now the question becomes whether it is an audible difference or not--but it is certain that the increased resolution of 24 bit exists throughout the whole dynamic range.

 

[sejarzo]

Quote:

Originally Posted by HFat /img/forum/go_quote.gif I don't want to hear peaks at 97dB or above except when the surroundings are very noisy and even then it's something I regret having done but I agree that this is a personal choice.

I don't know what sort of music you prefer, but much live orchestral music peaks well above 100 dB out in the house and is louder on stage. Such peaks aren't going to hurt your ears by any means.

 

[nick_charles]

Quote:

Originally Posted by sejarzo /img/forum/go_quote.gif I don't know what sort of music you prefer, but much live orchestral music peaks well above 100 dB out in the house and is louder on stage. Such peaks aren't going to hurt your ears by any means.

But the noise floor in your home is unlikely to be below 25db so to render a peak at 97db above the noise floor the actual overall sound level peak will be 122db, which is loud by any standards... Your orchestra will be operating in a hall which probably is just as noisy , full of noisy humans breathing, fans, and what have you.

 

[HFat]

Quote:

Originally Posted by sejarzo /img/forum/go_quote.gif Not really, seems to me that you are equating the resolution of a 0.25V change quantified from 2^22 to 2^21 as being equivalent to it being quantified between 2^14 to 2^13 ...

Yeah, I was quite sloppy. I should not have answered your post carelessly but now that the deed is done...
I was actually thinking in terms of individual bits and their positions rather than in terms of values. With all the talk of padding 16 bits data with zeros to get 24 bit data, I assumed the most significant bits to code much bigger changes in amplitude than the extra eight so that the first bits code the same amplitude change at every bit depth (not that you're saying anything else). I was trying to say that the extra "steps" you get would all have amplitudes which are (way) below the noise floor in most cases... which doesn't address your point really: I get the impression that you're actually saying that it may be possible to hear such "steps" in spite of the noise. I don't see how that would be possible if the noise was present at all frequencies but that's a different issue that the one you brought up I think. In other words, it's my turn to wonder exactly what your point was.

And you guessed it right: I don't listen to much (if at all) orchestral music. I lean towards chamber and away from the loudest instruments as far as classical is concerned. I don't like a bombastic dynamic range in music although I agree it fits movies well.

 

[HFat]

Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif But the noise floor in your home is unlikely to be below 25db so to render a peak at 97db above the noise floor...

We were talking about 97dB to zero (so to speak, what's the proper way to say that?), not relative to the noise.

There may be some confusion about what noise is or what noise does exactly among the least knowledgeable posters in this thread (that would include me).

 

[Tarkovsky]

It's been said that the volume levels would have to be deafening for 24 bit to make a difference, but this assumes you're not wearing hearing protection.
But wait that's silly you say, you've just lost you're noise floor by effectively lowering the whole outputs volume relative to your ears. Yes. But your ears aren't the only way you perceive sound.
If you're into D n B you'll know that a large part of the music is infrasound. So, in short I suppose we do have a use for 24 bit playback.
Some might not strictly describe this as music though. And I think you'd have a hard time ensuring everyone was wearing ear protection. It would incredible though.

 

[bigshot]

Just for perspective, 96dB is somewhere around the volume of a leaf blower. 120dB is the level of a jackhammer. You aren't going to have this kind of volume in your living room, even for short bursts.

See ya
Steve

 

[jenghanHsieh]

Quote:

Originally Posted by sejarzo /img/forum/go_quote.gif Punch it into a calculator and work it out for yourself. A 24 bit system does indeed provide much greater resolution of amplitude, and it really cannot be any other way... Now the question becomes whether it is an audible difference or not--but it is certain that the increased resolution of 24 bit exists throughout the whole dynamic range.

After reading this interesting thread and some related articles.

I think the 'perfect' means :

"A system with perfect precision, and good enough accuracy for the design purpose"

( see Error - Wikipedia for the definition of precision and accuracy ).

For Digital Audio,

[1]
In theory, the resulting amplitude in every given time slice can be 100% precise, independent of the bit-depth we choose. (with proper dithering ?)

[2]
The accuracy may vary according to the bit-depth we used, but it does not matter once we have high-enough accuracy for our purpose.

[3]
Less accuracy = higher noise floor, and, for playback, 16bits / 24bits is all about pushing down the noise floor.


If a system can hit the bull's-eye every single time, it is a perfect system.
It does not need to hit "EXACT THE SAME POINT EVERYTIME" to be perfect.

 

[Tarkovsky]

Yes but in a rave the size of a small airport it could come in handy!

 

[gregorio]

sejarzo - have a look at the graphs on the link below:

Quantization Error - DiracDelta Science & Engineering Encyclopedia

These graphs show exactly what you are describing and 20 years ago would represent what you would hear. However today, the dithering quantizer takes all those quantization errors and converts them into noise. So you don't get a stepped output, as displayed by the blue lines on the graphs, you get a perfectly linear output (at any bit depth), as displayed by the grey lines on the graphs. Look at the 4 graphs, what is the difference between the higher bit representations and the lower bit ones? the difference is that the higher bit depth has fewer quantization errors, and to a dithering quantizer this does not mean a more linear output but a lower noise floor as there are fewer errors requiring conversion into noise.

Remember, the noise floor isn't the end of audio, we can commonly hear signals substantially below the noise floor. Also remember that the dB scale is not linear but logarithmic. Double the dynamic range of 96dB is not 192dB, it's 99dB! So the difference between 16bit (96dB) and 24bit (144dB) is absolutely massive, thousands of times greater.

When using 24bit we are not talking about having to crank up your amp a little to hear the lower noise floor. We are talking about an incomprehensible dynamic range. In theory, 24bit is capable of capturing (above the noise floor) the sound pressure level created when two hydrogen atoms collide. If you crank your amp up high enough so that you could hear two hydrogen atoms colliding, when you get a fortissimo chord from the orchestra, you are going to wake up in hospital (or not at all!). Of course in practice, no microphone, amp or any other part of the signal chain is capable of anywhere near being able to capture or playback this dynamic range, which is why 24bit it totally superfluous as a playback medium.

The problem of high SPLs and hearing damage for orchestral musicians has been recognised for a long time. Particularly for viola and cello players who may be sitting directly in front of the trumpets and trombones. Ear plugs which reduce SPLs linearly are now routinely used by many orchestral musicians, at least during rehearsal if not for performance.

 

[HFat]

Quote:

Originally Posted by gregorio /img/forum/go_quote.gif These graphs show exactly what you are describing and 20 years ago would represent what you would hear. However today, the dithering quantizer takes all those quantization errors and converts them into noise.

I don't remember hearing "steps" back then but then again I can't hear the 16 bit quantization noise in normal conditions now (by a long shot).
I thought that the fact that I can't hear the noise implies that I couldn't hear the errors if they were left alone anyway... right or wrong?

Quote:

Originally Posted by gregorio /img/forum/go_quote.gif Remember, the noise floor isn't the end of audio, we can commonly hear signals substantially below the noise floor.

Even if the noise is white? Wouldn't it interfere with the faint signal, drowning it in randomness?

 

[nick_charles]

Quote:

Originally Posted by gregorio /img/forum/go_quote.gif Remember, the noise floor isn't the end of audio, we can commonly hear signals substantially below the noise floor. Also remember that the dB scale is not linear but logarithmic. Double the dynamic range of 96dB is not 192dB, it's 99dB! So the difference between 16bit (96dB) and 24bit (144dB) is absolutely massive, thousands of times greater.

Hang on, there is the doubling with 3db and doubling with 6db thing that are context dependent. Each bit in a digital PCM system doubles the dynamic range i.e a 16 bit system has 65536 levels and a 17 bit system has 131072 levels. And each bit gives you approximately 6.02db extra dynamic range. So a 24 bit system that has 144db dynamic range will have 16777216 levels which is 256 times more than 16 bits. It is still massively bigger, but not by thousands.

 

[sejarzo]

Quote:

Originally Posted by gregorio /img/forum/go_quote.gif sejarzo - have a look at the graphs on the link below: Quantization Error - DiracDelta Science & Engineering Encyclopedia These graphs show exactly what you are describing and 20 years ago would represent what you would hear. However today, the dithering quantizer takes all those quantization errors and converts them into noise. So you don't get a stepped output, as displayed by the blue lines on the graphs, you get a perfectly linear output (at any bit depth), as displayed by the grey lines on the graphs.

Nice graphs, but they represent ADC performance, not DAC performance. All of the X axes are labled as "Analogue input". So they don't show me "exactly" what I am hearing.

Please direct me to a similar reference that describes the difference in DAC operation, if you would, that shows the difference in analog output between 16 bit and 24 bit DAC operation when fed true 16 bit and 24 bit streams, respectively. I'm not a recording engineer, just a listener....and I have a collection of mostly 16 bit music, but also a number of 24 bit files.


Quote:

Originally Posted by gregorio /img/forum/go_quote.gif When using 24bit we are not talking about having to crank up your amp a little to hear the lower noise floor. We are talking about an incomprehensible dynamic range. In theory, 24bit is capable of capturing (above the noise floor) the sound pressure level created when two hydrogen atoms collide. If you crank your amp up high enough so that you could hear two hydrogen atoms colliding, when you get a fortissimo chord from the orchestra, you are going to wake up in hospital (or not at all!). Of course in practice, no microphone, amp or any other part of the signal chain is capable of anywhere near being able to capture or playback this dynamic range, which is why 24bit it totally superfluous as a playback medium.

So you are saying that there is no audible benefit at all to more accurate quanitization of signal levels in the say, -10 dBFS down to -40 dBFS range?

Quote:

Originally Posted by gregorio /img/forum/go_quote.gif The problem of high SPLs and hearing damage for orchestral musicians has been recognised for a long time. Particularly for viola and cello players who may be sitting directly in front of the trumpets and trombones. Ear plugs which reduce SPLs linearly are now routinely used by many orchestral musicians, at least during rehearsal if not for performance.

I certainly agree with that statement.

My point is that too many people on this forum are utterly confused re the difference between dynamic range of a recording medium and resulting SPL on playback. There is no strict connection; people seem to forget that dB is always a relative measurement rather than an absolute one.

Because a CD has a 96 dB range doesn't mean that a 0 dBFS peak will be harmful to your hearing, it all depends on how one sets playback volume.

A sine wave recorded at only 8 bit resolution at 0 dBFS and played back at 110 dB SPL would harm your hearing just as much as one recorded at 16 bit resolution, 0 dBFS, at the same SPL.

To me, 85 dB SPL (C weighted) pink noise sounds pretty loud during setting levels on a home theater system. But listening to wind ensemble music at an average level of 75 db SPL, at a lower volume setting on the processor than required to get the 85 dB SPL calibration during set up, includes peaks around 100 dB.....and I don't worry about that being harmful at all.

 

[grawk]

The only way you'll be able to hear a benefit to 24 bit is if you turn the volume up loud enough to be beyond the range of 16 bit. That means 96dB at a minimum, and 96dB above noise floor ideally. You definitely won't see a benefit between -10 and -40. Those bits are the same.

24 bit has and probably will remain primarily beneficial at record time, because it allows levels to be set high enough to capture dynamics while allowing enough headroom for unexpected or transient peaks. That way compression, if used at all, won't need to be applied at capture, but can be applied as needed or desired in post.

16bit exceeds the SNR of most gear, and certainly exceeds the distance between noise floor and the pain threshhold of human hearing.

 

[bigshot]

As I've been told, there is some added resolution in very low volume areas that overlap the edge of coverage between 16 and 24 bit. In normal listening, this added resolution would be below your ability to hear the difference, but it can make a difference in a mix where you might be boosting the level of something quiet to make it audible in the mix. I don't know all the gobbledegook principles behind why that is, just the practical application.

See ya
Steve